关键词: 正交邻域保持嵌入, 流形学习, 连续隐Markov模型, 经验模式分解, 故障诊断

Abstract: To realize automation and high accuracy of rotating machinery fault diagnosis, a model diagnosis method for rotating machinery was proposed based on Orthogonal Neighborhood Preserving Embedding (ONPE) and Continuous Hidden Markov Model (CHMM). Firstly, the fault vibration signals of moving parts were decomposed by EMD and Shannon entropy was constructed to obtain high-dimensional eigenvectors. Then, the high-dimensional eigenvectors were compressed and simplified as the low-dimensional eigenvectors by ONPE. Finally, the low-dimensional eigenvectors were put into CHMM for fault pattern recognition. Fault diagnosis example of deep groove ball bearings proved the effectiveness of this new method.

Key words: orthogonal neighborhood preserving embedding, manifold learning, continuous hidden Markov model, empirical mode decomposition, fault diagnosis